[openstack-dev] [neutron][lbaas] Can entity calls be made to driver when entities get associated/disassociated with root entity?

Jain, Vivek VIVEKJAIN at ebay.com
Tue Feb 3 15:59:29 UTC 2015


Hi Vijay,
My understanding was that each vendor will have different behavior for entity creation. LBaaS apis will mark each entity as PENDING_CREATE (?) initially and its up to specific vendor driver whether to mark entities as DEFERRED or actually CREATE them on LB. Vendor logic can decide the action based on whether hierarchy/criteria is met per their LB prerequisite.

Thanks,
Vivek

From: Vijay Venkatachalam <Vijay.Venkatachalam at citrix.com<mailto:Vijay.Venkatachalam at citrix.com>>
Reply-To: "OpenStack Development Mailing List (not for usage questions)" <openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>>
Date: Tuesday, February 3, 2015 at 5:13 AM
To: "OpenStack Development Mailing List (openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>)" <openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>>
Subject: [openstack-dev] [neutron][lbaas] Can entity calls be made to driver when entities get associated/disassociated with root entity?

Hi:

In OpenStack neutron lbaas implementation, when entities are created/updated by the user, they might not be associated with the root entity, which is loadbalancer.
Since root entity has the driver information, the driver cannot be called by lbaas plugin during these operations by user.
Such entities are set in DEFFERED status until the entity is associated with root entity.
During this association operation (listener created with pool), the driver api is called for the current operation (listener create); and the driver is expected to perform the original operation (pool create) along with the current operation (listener create).
This leads to complex handling at the driver, I think it will be better for the lbaas plugin to call the original operation (pool create) driver API in addition to the current operation (listener create) API during the association operation.

That is the summary, please read on to understand the situation in detail.

Let’s take the example of pool create in driver.


a.       A pool create operation will not translate to a pool create api in the driver. There is a pool create in the driver API but that is never called today.

b.      When a listener is created with loadbalancer and pool, the driver’s listener create api is called and the driver is expected to create both pool and listener.

c.       When a listener is first created without loadbalancer but with a pool, the call does not reach driver. Later when the listener is updated with loadbalancer id,  the drivers listener update  API is called and the driver is expected to create both pool and listener.

d.  When a listener configured with pool and loadbalancer is updated with new pool id,  the driver’s listener update api is called. The driver is expected to delete the original pool that was associated, create the new pool and  also update the listener

As you can see this is leading to a quite a bit of handling in the driver code. This makes driver code complex.

How about handling this logic in lbaas plugin and it can call the “natural” functions that were deferred.

Whenever an entity is going from a DEFERRED to ACTIVE/CREATE status (through whichever workflow) the plugin can call the CREATE pool function of the driver.
Whenever an entity is going from an ACTIVE/CREATED to DEFERRED status (through whichever workflow) the plugin can call the DELETE pool function of the driver.

Thanks,
Vijay V.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20150203/4ef4e31d/attachment.html>


More information about the OpenStack-dev mailing list